A Practical Look at Downscaling, Bias Correction, …...• Methods to relate downscaled fields to...
Transcript of A Practical Look at Downscaling, Bias Correction, …...• Methods to relate downscaled fields to...
Building Resilience to a Changing Climate:
A Technical Training in Water Sector
Utility Decision Support
A Practical Look at Downscaling, Bias
Correction, and Translating Climate
Science into Hydrology
Julie Vano, National Center for Atmospheric Research
Classic “Top-down” Impacts Modeling Chain
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g. RCP8.5
e.g. CESM
e.g. BCSD e.g. Sac-SMA
Management/Operations
Model(s)
e.g. WEAP, SWMM
2
Classic “Top-down” Impacts Modeling Chain
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g. RCP8.5
e.g. CESM
e.g. BCSD e.g. Sac-SMA
Management/Operations
Model(s)
e.g. WEAP, SWMM
3
Classic “Top-down” Impacts Modeling Chain
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g. RCP8.5
e.g. CESM
e.g. BCSD e.g. Sac-SMA
Management/Operations
Model(s)
e.g. WEAP, SWMM
4
Classic “Top-down” Impacts Modeling Chain
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g. RCP8.5
e.g. CESM
e.g. BCSD e.g. Sac-SMA
Management/Operations
Model(s)
e.g. WEAP, SWMM
5
Why Do We Need to Downscale?
6
Global climate models:
• Coarse resolution of
topography
• Inaccurate simulation of
orographic precipitation,
temperature gradients,
cloud, snow, etc.
Regional climate
models:
• High resolution of
topography
• More accurate simulation
of local physics and
dynamics
Benefits of Downscaling
• Downscaling provides local-scale insight
• Impacts models need fine-scale and high-
temporal resolution climate inputs (e.g.,
precipitation, temperature, winds, radiation,
moisture)
• Downscaling can correct for certain biases of
global climate models
7
Types of Downscaling: Dynamical
8
• Uses a high-resolution regional climate
model (e.g., WRF) to simulate local
dynamics over the area of interest
• Global model output is applied along the
boundaries and as initial conditions
• Computationally expensive, time and
supercomputers (usually) required
9
Dynamical Downscaling Output
• Uses statistical relationships that relate
coarse to fine resolution from historical
record
• Stationary statistical relationships then
applied to future global model output
• Output usually for subset variables
(temperature, precipitation)
• Computationally cheap, quick and can
be done anywhere
• Statistical relationships do an excellent
job reproducing historical data
Types of Downscaling: Statistical
10
Example: Bias correction with
spatial disaggregation (BCSD)
Tradeoffs Between
Dynamical and Statistical Downscaling
• Represents physical processes
• No stationarity assumptions
• Physically consistent across variables
• Computationally expensive
• Data set availability is limited
• Introduces need for additional ensembles
• Produces climate change signals that still must analyzed for credibility
Dynamical
11
• Computationally tractable for
large GCM ensembles
• Large high-resolution data sets
publicly available
• Consistent with observations
• May not represent climate
change signal correctly (often is
effectively just interpolated GCM
signal)
• Statistical nature often introduces
artifacts
Statistical
A Continuum of Downscaling Options
• Dynamical downscaling using state-of-the-art RCMs e.g., Water Research and Forecasting model
• ”Hybrid”
• (dynamical + statistical) downscaling
– Build statistical emulator of RCM using limited set of dynamical
runs
• Physically based intermediate-complexity models
– e.g. Linear Orographic Precipitation model
• Statistical downscaling based on GCM dynamics (water vapor, wind,
convective potential, etc.)
– Regression-based, analog, pattern scaling, etc.
• Methods to relate downscaled fields to synoptic scale atmospheric
predictorsweather typing, etc.
• Statistical downscaling based on rescaling GCM outputs e.g., BCSD, LOCA, BCCA, linear regression, and more
incre
asin
g p
hysic
al re
pre
senta
tion
12
• ”Hybrid” (dynamical + statistical) downscaling
e.g., build statistical emulator using limited set of dynamical runs
• Physically-based intermediate-complexity atmospheric models
e.g., Linear Orographic Precipitation model
• Statistical downscaling based on GCM dynamics (water vapor, wind,
convective potential, etc.)
e.g., regression-based, analog, pattern scaling
• Methods to relate downscaled fields to synoptic scale atmospheric
predictors
e.g., self-organized maps, weather typing
13
Simulations in the Northwest
Winter temperatures
HadRM3P NARR Differences
from OSU’s Weather@home project, 1979-2009
Regional Climate Simulations with Crowdsourced Computing, Mote et al. BAMS 2015NARR = North American Regional Reanalysis; HadRM3P= high-resolution, regional configuration of HadAM3
Simulations in the Northwest
Rupp et al., HESS, 2016
Change in
ensemble mean
temperature,
1985-2014 to
2030-59 for
RCP 4.5
Winter
Summer
Autumn
Spring
Simulations in the Northwest
Rupp et al., HESS, 2016
Temperature
changes at
47.75ºN, 1985-
2014 to 2030-59
for RCP 4.5
16
Simulations in the Northwest
Li et al., in prep, figure courtesy of Phil Mote
Change in
ensemble mean
precipitation,
1985-2014 to
2030-59 for
RCP 4.5
Winter
Summer Autumn
Spring
17
Simulations in the NorthwestGCM Change ICAR Change
ICAR
from NCAR’s Intermediate Complexity Atmospheric Research Model, Gutmann et al. 2016More at: https://ncar.github.io/hydrology/models/
Questions to Help Determine an
Appropriate Downscaling Technique
18
• How large is the area of interest?
• Where is it?
• What is the impact of interest?
• When in the future?
• Does the sequencing of events matter?
• What type of climate change uncertainty is
important?
• What is available?
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g., RCP8.5
e.g., CESM
e.g., BCSD e.g. Sac-SMA
Management/Operations
Model(s)
e.g., WEAP, SWMM
19
Classic “Top-down” Impacts Modeling Chain
What type of models do you
use to track water in your
system?
2020
What we have: precipitation,
temperature, other atmospheric
values
Why Do We Need Hydrology Models?
21
What we have: precipitation,
temperature, other atmospheric
values
What we would like: streamflow
(highs, lows), water demand from
vegetation, water temperature
Why Do We Need Hydrology Models?
22
What we have: precipitation,
temperature, other atmospheric
values
What we would like: streamflow
(highs, lows), water demand from
vegetation, water temperature
Hydrology models represent
energy and water fluxes in
watersheds, combine
measurements and physical
processes to encapsulate our
understanding.
Important in filling gaps since
measurements are not available
in most places.
Why Do We Need Hydrology Models?
23
Portland Water Bureau• Land surface values from GCMs
measures not helpful
• Worked with University of
Washington to select and set up
in-house hydrologic model
• Model allows PWB to understand
how changes in streamflow affect
future supply conditions
• Included in Supply System
Master Plan
Modeling Benefits
24
• Models built to represent many landscapes, processes, spatial configurations+
• May miss key elements• Snow redistribution
• Groundwater interactions
• Important to be a savvy user
25
Modeling Cautions
City of Portland
from Cedar River Watershed Education Center
from Portland Water Bureau?from Bureau of Environmental Services, City of Portland
Hydrologic Model Spatial Structures
26
Hydrologic Model Process Structure
Conservation equations
Water
Energy
Physical processes
XXX Model options
Evapo-transpiration
Infiltration
Surface runoff
SolverCanopy storage
Aquifer storage
Snow temperature
Snow Unloading
Canopy interception
Canopy evaporation
Water table (TOPMODEL)Xinanjiang
(VIC)
Rooting profile
Green-AmptDarcy
Frozen ground
Richards’Gravity drainage
Multi-domain
Boussinesq
Conceptual aquifer
Instant outflow
Gravity drainage
Capacity limited
Wetted area
Soil water characteristics
Explicit overland flow
Atmospheric stability
Canopy radiation
Net energy fluxes
Beer’s Law
2-stream vis+nir
2-stream broadband
Kinematic
Liquid drainage
Linear above
threshold
Soil Stress function
Ball-Berry
Snow drifting
Louis
Obukhov
Melt drip Linear reservoir
Topographic drift factors
Blowing snowmodel
Snowstorage
Soil water content
Canopy temperature
Soil temperature
Phase change
Horizontal redistribution
Water flow through snow
Canopy turbulence
Supercooled liquid water
K-theory
L-theory
Vertical redistribution
Clark et al. (WRR 2015)27
Looking under
the hood…
Hydrologic Model Construction
28
Conservation equations
Water
Energy
SolverCanopy storage
Aquifer storage
Snow temperature
Snowstorage
Soil water content
Canopy temperature
Soil temperature
Clark et al. (WRR 2015)
Conservation equations, the order they
are solved and time step matter
29
Conservation equations
Water
Energy
Physical processes
Evapo-transpiration
Infiltration
Surface runoff
SolverCanopy storage
Aquifer storage
Snow temperature
Snow Unloading
Canopy interception
Canopy evaporation
Atmospheric stability
Canopy radiation
Net energy fluxes
Liquid drainage
Snow drifting
Snowstorage
Soil water content
Canopy temperature
Soil temperature
Phase change
Horizontal redistribution
Water flow through snow
Canopy turbulence
Supercooled liquid water
Vertical redistribution
Clark et al. (WRR 2015)
Equations to calculate model fluxes
(e.g., evapotranspiration)
Hydrologic Model Construction
30
Conservation equations
Water
Energy
Physical processes
XXX Model options
Evapo-transpiration
Infiltration
Surface runoff
SolverCanopy storage
Aquifer storage
Snow temperature
Snow Unloading
Canopy interception
Canopy evaporation
Water table (TOPMODEL)Xinanjiang
(VIC)
Rooting profile
Green-AmptDarcy
Frozen ground
Richards’Gravity drainage
Multi-domain
Boussinesq
Conceptual aquifer
Instant outflow
Gravity drainage
Capacity limited
Wetted area
Soil water characteristics
Explicit overland flow
Atmospheric stability
Canopy radiation
Net energy fluxes
Beer’s Law
2-stream vis+nir
2-stream broadband
Kinematic
Liquid drainage
Linear above
threshold
Soil Stress function
Ball-Berry
Snow drifting
Louis
Obukhov
Melt drip Linear reservoir
Topographic drift factors
Blowing snowmodel
Snowstorage
Soil water content
Canopy temperature
Soil temperature
Phase change
Horizontal redistribution
Water flow through snow
Canopy turbulence
Supercooled liquid water
K-theory
L-theory
Vertical redistribution
Clark et al. (WRR 2015)
Hydrologic Process Flexibility
Model Parameters
• Parameters can represent real-world vegetation, soil type
• Calibrated parameters can compensate for other errors
• Compensating errors can respond differently to climate change
• Check robustness by exploring multiple parameter sets
31
http://usnvc.org/overview/
Vegetation, Soil type, …
31
32
Simulations in the Northwest
RMJOC-II: Predicting the Hydrologic Response of the
Columbia River to Climate Change
Project at UW and OSU
Data available at:
hydro.washington.edu/CRCC
Columbia River at The Dalles
33
Simulations in the Northwest
VIC-ORNL
VIC-UWPRMS
VIC-NCAR
Columbia River at The Dalles
*comparisons made prior to bias correction, pre-released version of dataset for illustration only
What Do Models Tell Us?
• Many responses to climate change are “obvious” but some are not
• Hydrology-climate interactions not always linear• Rain-on-snow events
• Slower snow melt in a warmer world
• Tipping points can be hard to detect
• Models encapsulate our understanding of the system, but far from perfect
34
GCM Initial
Conditions
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
Method(s)
Hydrologic
Model
Structure(s)
Hydrologic
Model
Parameter(s)
scenarios
ens. members
models
Combined uncertainty
projections
methodsmodels
calibration
Revealing Uncertainties
35
GCM Initial
Conditions
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
Method(s)
HydrologicModel
Structure(s)
Hydrologic
Model
Parameter(s)methods
scenarios
ens. members
models
models
calibration
Combined uncertainty
projections
36
Key Uncertainties from:
• Human activities
• Physical processes
• Natural variability
Revealing Uncertainties
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g., RCP8.5
e.g. ,CESM
e.g., BCSD e.g., Sac-SMA
Management/Operations
Model(s)
e.g., WEAP, SWMM
37
Classic “Top-down” Impacts Modeling Chain
Classic “Top-down” Impacts Modeling Chain
Emissions
Scenario(s)
Global Climate
Model(s)
Downscaling
method(s)Hydrologic
Model(s)
Decision
e.g., RCP8.5
e.g., CESM
e.g., BCSD e.g., Sac-SMA
Management/Operations
Model(s)
e.g., WEAP, SWMM
38
Revised
Paleoclimate studies
Clark et al. 2016; connect
models in a chain
Scenario studies
Brown et al., WRR, 2016; explore system vulnerabilities
with perturbations
Climate-informed
vulnerability analysis
80% confidence intervals
Vano et al., BAMS, 2016; generate timeseries using reconstructions of the distant past
Stochastic hydrology
Bras and Rodriguez-Iturbe, 1985; generate synthetic timeseries using
statics from the past
and
others…
39
Do Be Aware of Multiple Ways to Evaluate Future Changes
• Certain models and methods are
more appropriate
• Certain spatial and temporal scales
are more appropriate for certain
questions
• Realize some questions may not
be possible to answer with current
knowledge
• Finer resolution in space and time
is not necessarily better• Higher Resolution ≠ Higher Accuracy
Be a savvy consumer and
remember…Different: GCMs, emission scenarios,
spatial resolution, hydrology, +
Figure from Vano et al., BAMS, January 201440
Don’t Treat All Future Projections or Methods Equally
“The accuracy of streamflow simulations in natural catchments will always be
limited by simplified model representations of the real world as well as the
availability and quality of hydrologic measurements.” (Clark et al., WRR, 2008)
• Don’t expect perfect results,
• Not prediction, but a tool to test how system responds
(what if scenarios)
• BUT we can make better choices…
• Seek simple yet defensible (don’t need a Cadillac)
• Be aware of models shortcomings (know the warts)
41
No Model is Perfect
41
• Hydrology focused Green Data Oasis (GDO) portal
– BCSD (12km), LOCA (6km)
– VIC streamflow
• Dynamical– NARCCAP (50km),
– CORDEX (limited 25km)
– Others over regional domains or limited time periods
• USGS GeoDataPortal– Collection of different archives
• Northwest Knowledge Network– MACA (6 km, 4 km)
• RMJOCII, Columbia River Climate Change– BCSD, MACA
– VIC, PRMS streamflow
• Many others (NASA NEX, ARRM)
What Data are Available Now?
42
What Resources are Available?• WUCA products
– PUMA project examples
– www.wucaonline.org
• Federal Agency Guidance– Bureau of Reclamation
– U.S. Army Corps of Engineers
– Environmental Protection Agency
– U.S. Climate Resilience Toolkit
• Professional Societies– American Society of Civil Engineers
• Regional Boundary Organizations– Climate Impacts Research Consortium
(NOAA RISA) at OSU
– Climate Impacts Group, UW
• Dos and Don’ts Guidelines from NCAR– Reviews other guidance
– www.ncar.github.io/dos_and_donts
• Many others, including each other
Climate Change Study Choices
44Clark et al. 2016
• Approach type (e.g. scenarios, paleo, vulnerability analysis):
• Emission scenarios used:• GCMs used: • Number of initial conditions for each GCM
used: • Downscaling methods used:• Hydrologic models and parameter sets used:• Time period of interest (transient or delta):• Project timeline:• Impacts evaluated:• Results reported (ensembles, individual
simulations):
Key Takeaways• Downscaling and hydrology modeling provide local-
scale insights into possibilities projected by GCMs.
• There is a continuum of downscaling approaches that
span tradeoffs between computational efficiency and
methodological complexity.
• Some change signals are more certain than others.
• Some uncertainty is unavoidable.
– Representation of uncertainties is hard but necessary.
– Uncertainties have always been there; just understanding them
now.
– Previous studies may be over-confident.
45
Key Takeaways
• Research underway to develop ways to select
representative set of scenarios useful for water
resources planning.
• It is critical to understand important processes and
uncertainties in your system.
• Models are tools that can be useful, if used
appropriately. Be a savvy consumer.
• Consult local experts and national resources,
e.g., OSU, UW, NCAR
https://ncar.github.io/dos_and_donts
46